How Primoraly works
Four steps to use it — and below, exactly how the measurement is made, because you should not have to trust a score you cannot interrogate.
Free · 30 seconds · no account needed
What using it looks like
Add your business
Enter your website and keywords. Local businesses set a service area and language; the questions are asked the way your customers ask them.
We scan & check
Your site is scanned for how readable it is to AI systems, and your keywords are checked live across ChatGPT, Claude, Perplexity and Gemini.
Read the answers
Every answer is stored verbatim: whether you were named, where, in what words, and which competitors appeared instead of you.
Change, re-check, compare
Make changes, run the same check again — same questions, thanks to seeding — and see whether the answers moved. Scheduled reports keep it in front of you.
Methodology
How the measurement is made
Different AI-visibility tools can give different answers to the same prompt — a fair criticism of the whole category. The honest response is to show our sampling, so here it is.
12 answers per keyword
Each keyword check asks three different questions of each of the four assistants — twelve answers in total. The three questions are deliberately different kinds: a ranked-list question (“top ten…”), an alternatives question (“alternatives to X”), and a recommendation question. Visibility depends on which question gets asked; sampling only one kind measures only one kind.
The same question every time
Question generation is seeded per business and keyword, so a check this week and a check next month ask the identical wording. Without that, a moving trend line could just mean the question was re-worded. When you compare runs on Primoraly, you are comparing the answers, not the phrasing.
Live models, not pinned ones
We ask whatever each provider currently serves its users — rolling aliases where providers publish them — because the thing being measured is what a customer asking today would be told. The trade-off is honest: last month's run may have asked a different model version, which is exactly what a real customer would have experienced.
Whole-word mention matching
A business is only counted as mentioned when its name or domain appears as an actual reference, not as a substring. A company whose domain is “on.com” does not get credit for the word “on”. Position is read from list structure when present.
Run-to-run variance, stated plainly
LLMs are not deterministic: the same question can name eight businesses one run and seven the next. We reduce the noise by fixing the questions and reading multiple answers per keyword, and we date every measurement. What we do not do is present a single answer as the truth — the trend across runs is the signal.
What we don't do
Primoraly measures. It does not edit your website, publish content for you, or promise to change what AIs say — and no honest tool can guarantee rankings in AI responses. What the measurement gives you is the gap: the exact questions where competitors are named and you are not, which is where the work is.